1 下载和安装

在ollama的官网下载mac m1版本的ollama

https://ollama.com/

最终获得如下所示的下载地址

https://github.com/ollama/ollama/releases/latest/download/Ollama.dmg

然后点击安装，然后测试

ollama list

2  运行deepseek r1

deepseek-r1:8b 比较适合笔记本部署，大约5G左右，当然也可以根据电脑配置选择其他版本。

ollama run deepseek-r1:8b

ollama将deepseek-r1:8b拉到本地后，就可以开始交互测试了。

3  api调用ollama model

1）初始设置

本地ollama端口11434

import requests

# 基础初始化设置
base_url = "http://localhost:11434/api"
headers = {
    "Content-Type": "application/json"
}

调用llm示例代码，可以按需指定llm，比如"deepseek-r1:8b"

2）流式生成文本补全

(Streaming Completion)

import requests
import json

# 基础初始化设置
base_url = "http://localhost:11434/api"
headers = {
    "Content-Type": "application/json"
}
def generate_completion_stream(prompt, model="deepseek-r1:8b"):
    url = f"{base_url}/generate"
    data = {
        "model": model,
        "prompt": prompt,
        "stream": True,
        "options": {
            "temperature": 0.7
        }
    }
    response = requests.post(url, headers=headers, json=data, stream=True)
    result = ""
    for line in response.iter_lines():
        if line:
            d = json.loads(line)
            resp = d["response"]
            result += str(resp)
    return result


# 示例调用
prompt = "什么是自然语言处理？"

stream_completion = generate_completion_stream(prompt)
print("流式生成文本补全:", stream_completion)

3）生成对话补全

(Chat Completion)

def generate_chat_completion(messages, model="qwen2.5:0.5b"):
    url = f"{base_url}/chat"
    data = {
        "model": model,
        "messages": messages,
        "stream": False,
        "options": {
            "temperature": 0.7
          },
    }
    response = requests.post(url, headers=headers, json=data)
    return response.json().get('message', {}).get('content', '')

# 示例调用
messages = [
    {"role": "user", "content": "什么是自然语言处理？"},
    {"role": "assistant", "content": "自然语言处理是人工智能的一个领域，专注于人与计算机之间的自然语言交互。"}
]
chat_response = generate_chat_completion(messages)
print("生成对话补全:", chat_response)

4）生成文本嵌入

(Generate Embeddings)

def generate_embeddings(text, model="qwen2.5:0.5b"):
    url = f"{base_url}/embed"
    data = {
        "model": model,
        "input": text
    }
    response = requests.post(url, headers=headers, json=data)
    return response.json()

# 示例调用
embeddings = generate_embeddings("什么是深度学习？")
print("生成文本嵌入:", embeddings)

5）列出本地模型

(List Local Models)

def list_local_models():
    url = f"{base_url}/tags"
    response = requests.get(url, headers=headers)
    return response.json().get('models', [])

# 示例调用
local_models = list_local_models()
print("列出本地模型:", local_models)

6）查看模型信息

(Show Model Information)

def show_model_info(model="qwen2.5:0.5b"):
    url = f"{base_url}/show"
    data = {"name": model}
    response = requests.post(url, headers=headers, json=data)
    return response.json()

# 示例调用
model_info = show_model_info()
print("模型信息:", model_info)

7）创建模型

(Create a Model)

def create_model(model_name="qwen2.5_custom", base_model="qwen2.5:0.5b"):
    url = f"{base_url}/create"
    data = {
        "name": model_name,
        "modelfile": f"FROM {base_model}\nSYSTEM You are a helpful assistant."
    }
    response = requests.post(url, headers=headers, json=data)
    return response.json()

# 示例调用
create_response = create_model()
print("创建模型:", create_response)

8）拉取模型

def pull_model(model_name="qwen2.5:0.5b"):
    url = f"{base_url}/api/pull"
    data = {"name": model_name}
    response = requests.post(url, headers=headers, json=data)
    return response.json()

# 示例调用
pull_response = pull_model()
print("拉取模型:", pull_response)

9）删除模型

def delete_model(model_name="qwen2.5_custom"):
    url = f"{base_url}/delete"
    data = {"name": model_name}
    response = requests.delete(url, headers=headers, json=data)
    return response.json()

# 示例调用
delete_response = delete_model()
print("删除模型:", delete_response)

相比llama.cpp，ollama集成各种工具，使用更简单。

使用openai接口访问ollama模型的例子见

如何用OpenAI SDK调用Ollama LLM_openai 连接ollama-CSDN博客

reference

----

ollama

https://github.com/ollama/ollama

在mac m1基于llama.cpp运行deepseek

https://blog.csdn.net/liliang199/article/details/149246699

如何用OpenAI SDK调用Ollama LLM

https://blog.csdn.net/liliang199/article/details/150111535